VINSFusion

VINS-Fusion is a real-time multi-sensor SLAM framework that fuses visual data from one or more cameras with inertial measurements from an IMU to estimate a vehicle’s ego-motion and, in many configurations, to build a map of the surrounding environment. It extends the capabilities of earlier visual-inertial systems by supporting multiple camera configurations and the integration of additional sensor data for improved accuracy and global consistency.

Key features include a tightly coupled, sliding-window nonlinear optimization backend, IMU pre-integration for efficient fuse with visual measurements, and robust feature tracking and data association. The system supports monocular, stereo, and RGB-D camera rigs and provides online calibration of sensor extrinsics and intrinsics. It is designed to operate in real time on common robotics platforms and is commonly deployed with ROS, offering modular components for sensor input, state estimation, optimization, and visualization.

VINS-Fusion is capable of incorporating non-visual sensors such as GNSS to improve global position estimates and drift reduction, and it can accommodate additional odometry information to enhance robustness on wheeled or aerial platforms. The framework emphasizes flexibility and portability, enabling researchers to adapt the pipeline to different hardware setups and sensor suites. Open-source and widely used in robotics and autonomous systems communities, VINS-Fusion serves as a reference implementation for robust visual-inertial navigation with multi-sensor fusion, supporting applications in drones, ground vehicles, and handheld devices.